Apple Reportedly Reduced Face ID Accuracy to Speed Up iPhone X Production

[teknishn]

It depends on the definition of "standards" is for this component.

For example, say FaceID was required to have an accuracy of "7" to meet standards, and Apple started out making them at "10", but "10" was not sustainable, so they dropped it to 7 or 8. That would be a drop, but still in their range of "meeting standards".

That happens quite a bit in mass production.

This^ Apple is known for these kind of insanely stringent standards. This is why their gear is generally regarded as the highest quality. We tend to take it for granted, but you just don't typically see iPhones with all the problems other vendors are facing, while at the same time producing 10s of millions more units at the high end. Take the dumpster fire that is the Pixel 2 phones right now as an example. I read another report months ago that, early in manufacturing, Samsung was having trouble meeting Apple's stringent requirements for the display. This is the same Samsung that uses OLED in all their top phones for years having trouble meeting Apple's quality requirements. Requirements that were so high, no other OLED supplier could compete for the business.

Apple builds in these buffers for this sort of reason. FaceID is going to be just fine folks.

 

[Tech for Kings]

They never comment on these supplier rumors. They usually don't even want to confirm who their suppliers are.

They surprisingly did comment this time.

 

[deanthedev]

Apple's production issues with the X are not rumor. Is this particular rumor true? I don't know. But Apple's X yields are going to underperform for the year. FACT.

Not rumor? Then please ink to the official statements from Apple or any suppliers confirming the production issues.

 

[ravenstar]

we would if they allowed reviews before pre-orders, best you can do is pre-order and return it if its not up to par.

this rumor describes exactly what happened to samsung tho lol, either their full of it or Apple really screwed up.

Nope, the best I can do is wait for it to be released, then look at the reviews and make a decision. There's absolutely no reason I need to be the first to have this device, and I very much doubt anyone else *must* have it on the first day of sale. Certainly if you feel returning it is an option, it's not a must have.

 

[MacsRuleOthersDrool]

Dear Apple,

Get it right. Then sell it. The yearly upgrade cycle is killing quality.

It was the Jobs-Era Apple that started the Yearly Upgrade treadmill...

 

[ravenstar]

Worst launch ever and royal **** up for Tom Cook from start to finish if true. Maybe it’s time to find a real CEO

Let's say Apple did this but you never heard about it. Would you say the same thing?

 

[SuperMatt]

They surprisingly did comment this time.

Yes I just saw that. I noticed they made no comments about any suppliers. They simply state that the reduction in Face ID accuracy is completely false.

 

[greeley]

Nope, the best I can do is wait for it to be released, then look at the reviews and make a decision. There's absolutely no reason I need to be the first to have this device, and I very much doubt anyone else *must* have it on the first day of sale. Certainly if you feel returning it is an option, it's not a must have.

you can try, nobody knows how back ordered these devices will be. you might be waiting for January for them to finally catch up with demand.

 

[Swift]

Apple has now denied it, per Twitter.

 

[Ankaa]

which is pretty telling, to be honest. Apple has been much more open regarding giving statements in the past years than before (like with the iTunes rumor).

Nice to see Apple actually commented and debunked it.

 

[MacsRuleOthersDrool]

Does this mean that a batch of phones have a better quality FaceID system than others?

Can see it now.... how can I tell if I have the good TrueDepth camera system, or the gimped/compromise one

Samsung/TSMC SoC, Qualcomm/Intel Modems, all over again....

ALL manufacturers that use "identical" components from multiple suppliers suffer the same issues: If it's not in "The Spec", then supplier-to-supplier variations of some other Spec are not considered a reason to disqualify that supplier; or, in the case of the Cell Modems, if some OTHER requirement (like CDMA) makes it so you MUST use a particular supplier (that ISN'T your "Primary Source"), then tough beans: You (the manufacturer) pretty much HAVE to make a "Variance" for whatever Spec(s) that are different in the "Alternate Source".

Disclosure: I have been involved for years in industrial control product design/development/production. This stuff happens ALL the time, to EVERY SINGLE MANUFACTURER.

You only HEAR about it when APPLE is involved...

 

[Kabeyun]

I think report states that Dot Projector modules were not meeting the mandated Face ID requirements in terms of accuracy where usable rate is only 20% of the total parts manufactured.

It is the amount of testing that involves many stiff parameters delay the production process (QC). By reducing the number of steps in the QC they are assured of far higher yields.

No, the article says nothing about failure of modules to meet accuracy requirements. The article discusses overall yields, not sacrificed detection accuracy. Nowhere will you find anything describing specific compromises in FaceID accuracy or how such compromises would address yield.

 

[VulchR]

Sweet Jeebus please stop with the glasses, contacts, and other unrelated "impediments" to Face ID. Uggh, still with the false negative. It's not a hill to die one, trust me. The false negative is going to vary greatly based on how your phone held in relation to face. Certain angle are going to be better for authentication and certain angles are going to work better for different people.


I'm not saying you're not a scientist. But if you are a scientist, I'm of the opinion you're not doing what a scientist should. I say that because you've opted not to do the one fundamental thing a good scientist does: research. A cursory Google search will provide you with information about how Face ID works and what data it's actually capturing. I am not a scientist, but I am most definitely willing to help you get started because your familiarity with neural network modeling (including knowing people who know stuff about unrelated stuff) isn't apropos or necessary.
Face ID doesn't care about your face per se. It cares about the math of your face: The TrueDepth camera captures accurate face data by projecting and analyzing over 30,000 invisible dots to create a depth map of your face and also captures an infrared image of your face. A portion of the A11 Bionic chip's neural engine — protected within the Secure Enclave — transforms the depth map and infrared image into a mathematical representation and compares that representation to the enrolled facial data.
So neither glasses nor contact lenses will change that math. The caveat there is polarized glasses. Contact lenses don't even factor. There's plenty of information out there regarding how Face ID works and about the state of facial recognition in general. As a scientist that should have been where you started. My opinion of course.

Let me put it this way: I am content to see how people get on with FaceID before I invest my money in a new phone. My reluctance is due to the difficult (damned near impossible) problems they are trying to solve in recognising faces by a 3-D map.

 

[Ramchi]

No, the article says nothing about failure of modules to meet accuracy requirements. The article discusses overall yields, not sacrificed detection accuracy. Nowhere will you find anything describing specific compromises in FaceID accuracy or how such compromises would address yield.

"That's because the components used to build the device's facial-recognition sensor are so complex that it's been nearly impossible to mass-produce."

"During production of the projector modules, Bloomberg reports that only 20 percent of them were actually usable. The results prompted manufacturers to dramatically slow down their production lines in an attempt to improve the yield on the components. The report says that, crucially, Apple has "relaxed some of the specifications for Face ID."

Clearly QC procedures may be adjusted to meet the relaxed specifications to meet higher yields. Does it compromise Face ID accuracy or security, at this point no one probably knows.

 

[VulchR]

The white paper clearly answers the underlying question with enough detail enough detail: On sudden change of appearance and failure of recognition and then successful authorization, the phone will use the capture from the previous failure in order to adapt to that change by temporarily adding it to the samples. None of this "confuse the neural net" bull you're spouting even makes sense: All of your questions are meant to poke at some irrelevant bits to sound very smart, but you haven't even read or comprehended the most minimal of materials out there in the relevant topic.

OK, so this boils down to learning rate in the network. If it is too high, the network is liable to learn 'noise', for wont of a better word. If it is too slow, then it won't adapt quickly enough enough to people taking glasses on and off. Worse case scenario: the network represents the average of glasses on and off, making template matching in either case problematical. At any rate, all I am saying is that Apple hasn't really released enough technical details to know how well FaceID will perform, but they claim it will perform well. Like I said, it'll probably work for most people most of the time, but since I wear glasses sometimes and contacts sometimes, I personally will wait until we have real-world user reports of their experiences with the phone.

 

[69Mustang]

Let me put it this way: I am content to see how people get on with FaceID before I invest my money in a new phone. My reluctance is due to the difficult (damned near impossible) problems they are trying to solve in recognising faces by a 3-D map.

I 100% agree with you. If I was interested in the X, it wouldn't get purchased until I was supremely confident the phone worked as advertised.

 

[T Coma]

This article seems to have rustled some jimmies.

Aye. Some jimmies more than others.

Assuming I can HTML correctly, of course.

 

[VulchR]

Aye. Some jimmies more than others.

View attachment 727300

Assuming I can HTML correctly, of course.

This is MR....

 

[Mike MA]

Their response is no response. Not a denial, just silence.

It’s probably no big deal tbh. FaceID was already much more secure than TouchID. It probably still is, despite this.

They responded.

 

[Sl1m]

Agree named sources are better than unnamed, but I wouldn't call Bloomberg "random." They are a legit news org.

From whom they got the information, was unnamed.

 

[bnekic]

Confirmation that it is indeed complete BS; statement from Apple:

View attachment 727288

Dude how did you know this?! https://techcrunch.com/2017/10/25/...f-face-id-to-meet-demand-is-completely-false/

 

[Kabeyun]

"That's because the components used to build the device's facial-recognition sensor are so complex that it's been nearly impossible to mass-produce."

"During production of the projector modules, Bloomberg reports that only 20 percent of them were actually usable. The results prompted manufacturers to dramatically slow down their production lines in an attempt to improve the yield on the components. The report says that, crucially, Apple has "relaxed some of the specifications for Face ID."

Clearly QC procedures may be adjusted to meet the relaxed specifications to meet higher yields. Does it compromise Face ID accuracy or security, at this point no one probably knows.

No idea what “relaxed specifications” means. As you said, we don’t know the effect, if any, addressing yield has had on on accuracy, because the article doesn’t say. Maybe yield efforts have compromised durability. Maybe yield efforts have compromised reliability. We don’t know. Therefore it’s bad journalism for a headline to blare an unsupported claim like, “Apple reportedly reduced accuracy.”

 

[DoctorTech]

A question for you.................

How many days do you think it will be, after the public get sent the new iPhoneX for there to be a report and/or a YouTube video, showing two different people, being able to unlock the one iPhoneX with Face ID using either of their faces?

Faces that, to "Us Humans" we can see they look different people, but the iPhoneX get fooled.

How long?

1 day
5 days
30 days
60 days
365 days
Never?

Any thoughts?

Apple claims the odds of a stranger being able to unlock a phone with FaceID is approximately 1 in 1,000,000 (about 20 times better than the 1 in 50,000 chance of a stranger being able to unlock an iPhone with TouchID). Apple also stated the odds of a different user being able unlock FaceID goes up with "close relatives". We might very well see a video on YouTube in a couple of weeks with identical twins unlocking each other's iPhones but short of a twin, I really doubt we see a legitimate (i.e. non-staged) video like you described.

 

[1applerules1]

Quantity over Quality. Making it on time over Making it right. The new Apple way. Just look at iOS 11.

Nice comment, turns out Apple just proved you wrong in a response XD

 

[polterbyte]

They never comment on these supplier rumors. They usually don't even want to confirm who their suppliers are.

Except they just did. And, indeed, denied the rumor.
[doublepost=1508957223][/doublepost]

You have 14 days to return it. What's the point in waiting then? It doesn't work as you expected, return it for a full refund.

It would work like that if I lived in the USA. I live in Brazil and would have someone pre-order it for me and put it in a relative's hand, who's coming down here mid-november.

It's all moot, though: Apple just denied the whole "news".